I’ve been on a quest to enhance the Q&A capabilities within our team’s application, specifically focusing on reading domain-specific files. A few months ago, I experimented with OpenAI’s text-embedding-ada-002. My approach was to generate embeddings and then index each document. To retrieve answers, I computed similarities to identify the most relevant sections from the documents. However, this manual process proved to be cumbersome. What was even more discouraging was the accuracy of the results; they were far from satisfactory.

Recently, I came across promising developments in the Llama index and Langchain. Inspired by their use in financial reports AI analysis, I decided to experiment with Llama index’s asynchronous operations. While the results were impressive on a local instance, the performance was sluggish. Furthermore, I encountered challenges when trying to integrate it into our factset.io platform. The asynchronous operations seemed to be causing frequent timeouts, making the integration problematic.

While researching Llama index and Langchain, I stumbled upon a feature that caught my attention: they both have bubbled Q&A widget, powered by Mendable ai. The ease of use and accuracy of this widget were remarkable. It led me to think: if leading AI startups are integrating Mendable AI, it surely warranted further exploration on my part.

So, I decided to delve deeper into Mendable AI’s offerings, and the experience was enlightening. The platform seamlessly handled many of the intricate technological processes in the background. With just a few lines of code,

from mendable import ChatApp #python -m pip install mendable-py first
my_docs_bot = ChatApp(api_key=MENDABLE_API_KEY)
# my_docs_bot.add("github", "https://github.com/znaixian/Research")
# Ask a question to Mendable
answer = my_docs_bot.query("what is it about china theme ETFs?")
print(answer)

I could effortlessly integrate a GitHub repository page into our team app, enabling us to pose questions about its content directly. What’s more, Mendable AI offers customization options, allowing users to train the system and adjust parameters like temperature, length, and precision of the output. So far, my exploration on Mendable AI has been satisfying, and I will continue exploring it.